chore(3p/git): Unvendor git and track patches instead r/2903

This was vendored a long time ago under the expectation that keeping
it in sync with cgit would be easier this way, but it has proven not
to be a big issue.

On the other hand, a vendored copy of git is an annoying maintenance
burden. It is much easier to rebase the single (dottime) patch that we
have.

This removes the vendored copy of git and instead passes the git
source code to cgit via `pkgs.srcOnly`, which includes the applied
patch so that cgit can continue rendering dottime.

Change-Id: If31f62dea7ce688fd1b9050204e9378019775f2b

1 files changed, 0 insertions, 96 deletions

diff --git a/third_party/git/Documentation/RelNotes/1.7.1.1.txt b/third_party/git/Documentation/RelNotes/1.7.1.1.txt
deleted file mode 100644
index 3f6b3148a309..000000000000
--- a/third_party/git/Documentation/RelNotes/1.7.1.1.txt
+++ /dev/null
@@ -1,96 +0,0 @@
-Git v1.7.1.1 Release Notes
-==========================
-
-Fixes since v1.7.1
-------------------
-
- * Authentication over http transport can now be made lazily, in that the
-   request can first go to a URL without username, get a 401 response and
-   then the client will ask for the username to use.
-
- * We used to mistakenly think "../work" is a subdirectory of the current
-   directory when we are in "../work-xyz".
-
- * The attribute mechanism now allows an entry that uses an attribute
-   macro that set/unset one attribute, immediately followed by an
-   overriding setting; this makes attribute macros much easier to use.
-
- * We didn't recognize timezone "Z" as a synonym for "UTC" (75b37e70).
-
- * In 1.7.0, read-tree and user commands that use the mechanism such as
-   checkout and merge were fixed to handle switching between branches one
-   of which has a file while the other has a directory at the same path
-   correctly even when there are some "confusing" pathnames in them.  But
-   the algorithm used for this fix was suboptimal and had a terrible
-   performance degradation especially in larger trees.
-
- * "git am -3" did not show diagnosis when the patch in the message was corrupt.
-
- * After "git apply --whitespace=fix" removed trailing blank lines in an
-   patch in a patch series, it failed to apply later patches that depend
-   on the presence of such blank lines.
-
- * "git bundle --stdin" segfaulted.
-
- * "git checkout" and "git rebase" overwrote paths that are marked "assume
-   unchanged".
-
- * "git commit --amend" on a commit with an invalid author-name line that
-   lacks the display name didn't work.
-
- * "git describe" did not tie-break tags that point at the same commit
-   correctly; newer ones are preferred by paying attention to the
-   tagger date now.
-
- * "git diff" used to tell underlying xdiff machinery to work very hard to
-   minimize the output, but this often was spending too many extra cycles
-   for very little gain.
-
- * "git diff --color" did not paint extended diff headers per line
-   (i.e. the coloring escape sequence didn't end at the end of line),
-   which confused "less -R".
-
- * "git fetch" over HTTP verifies the downloaded packfiles more robustly.
-
- * The memory usage by "git index-pack" (run during "git fetch" and "git
-   push") got leaner.
-
- * "GIT_DIR=foo.git git init --bare bar.git" created foo.git instead of bar.git.
-
- * "git log --abbrev=$num --format='%h' ignored --abbrev=$num.
-
- * "git ls-files ../out/side/cwd" refused to work.
-
- * "git merge --log" used to replace the custom message given by "-m" with
-   the shortlog, instead of appending to it.
-
- * "git notes copy" without any other argument segfaulted.
-
- * "git pull" accepted "--dry-run", gave it to underlying "git fetch" but
-   ignored the option itself, resulting in a bogus attempt to merge
-   unrelated commit.
-
- * "git rebase" did not faithfully reproduce a malformed author ident, that
-   is often seen in a repository converted from foreign SCMs.
-
- * "git reset --hard" started from a wrong directory and a working tree in
-   a nonstandard location is in use got confused.
-
- * "git send-email" lacked a way to specify the domainname used in the
-   EHLO/HELO exchange, causing rejected connection from picky servers.
-   It learned --smtp-domain option to solve this issue.
-
- * "git send-email" did not declare a content-transfer-encoding and
-   content-type even when its payload needs to be sent in 8-bit.
-
- * "git show -C -C" and other corner cases lost diff metainfo output
-   in 1.7.0.
-
- * "git stash" incorrectly lost paths in the working tree that were
-   previously removed from the index.
-
- * "git status" stopped refreshing the index by mistake in 1.7.1.
-
- * "git status" showed excess "hints" even when advice.statusHints is set to false.
-
-And other minor fixes and documentation updates.